The following patch fixes some testsuite regressions of LRA in
comparison with reload on PA-RISC. There are still few regressions
which I'll work on later.
The patch was bootstrapped on x86/x86-64.
Committed as rev. 186823.
2012-04-25 Vladimir Makarov <vmaka...@redhat.com>
* lra-constraints.c (get_op_mode): Take mode from operand first
for asm.
(match_reload): Use gen_lowpart_SUBREG for scalar int modes.
(curr_insn_transform): Unshare subregs for inout operands.
Index: lra-constraints.c
===================================================================
--- lra-constraints.c (revision 186357)
+++ lra-constraints.c (working copy)
@@ -797,16 +797,19 @@ static inline enum machine_mode
get_op_mode (int nop)
{
rtx *loc;
- enum machine_mode mode = curr_static_id->operand[nop].mode;
+ enum machine_mode mode;
+ bool md_first_p = asm_noperands (PATTERN (curr_insn)) < 0;
/* Take mode from the machine description first. */
- if (mode != VOIDmode)
+ if (md_first_p && (mode = curr_static_id->operand[nop].mode) != VOIDmode)
return mode;
loc = curr_id->operand_loc[nop];
/* Take mode from the operand second. */
mode = GET_MODE (*loc);
if (mode != VOIDmode)
return mode;
+ if (! md_first_p && (mode = curr_static_id->operand[nop].mode) != VOIDmode)
+ return mode;
/* Here is a very rare case. Take mode from the context. */
return find_mode (&PATTERN (curr_insn), VOIDmode, loc);
}
@@ -858,14 +861,20 @@ match_reload (signed char out, signed ch
reg = new_in_reg
= lra_create_new_reg_with_unique_value (inmode, in_rtx,
goal_class, "");
- new_out_reg = gen_lowpart_SUBREG (outmode, reg);
+ if (SCALAR_INT_MODE_P (inmode))
+ new_out_reg = gen_lowpart_SUBREG (outmode, reg);
+ else
+ new_out_reg = gen_rtx_SUBREG (outmode, reg, 0);
}
else
{
reg = new_out_reg
= lra_create_new_reg_with_unique_value (outmode, out_rtx,
goal_class, "");
- new_in_reg = gen_lowpart_SUBREG (inmode, new_out_reg);
+ if (SCALAR_INT_MODE_P (outmode))
+ new_in_reg = gen_lowpart_SUBREG (inmode, new_out_reg);
+ else
+ new_in_reg = gen_rtx_SUBREG (inmode, new_out_reg, 0);
}
bitmap_set_bit (&lra_matched_pseudos, REGNO (reg));
bitmap_set_bit (&lra_bound_pseudos, REGNO (reg));
@@ -3247,7 +3256,7 @@ curr_insn_transform (void)
&& type != OP_OUT)
{
push_to_sequence (before);
- lra_emit_move (new_reg, *loc);
+ lra_emit_move (new_reg, old);
before = get_insns ();
end_sequence ();
}
@@ -3257,7 +3266,12 @@ curr_insn_transform (void)
if (find_reg_note (curr_insn, REG_UNUSED, old) == NULL_RTX)
{
start_sequence ();
- lra_emit_move (old, new_reg);
+ /* We don't want sharing subregs as the pseudo can
+ get a memory and the memory can be processed
+ several times for eliminations. */
+ lra_emit_move (GET_CODE (old) == SUBREG && type == OP_INOUT
+ ? copy_rtx (old) : old,
+ new_reg);
emit_insn (after);
after = get_insns ();
end_sequence ();
